1 Highly variable species distribution models in a subarctic stream

2 metacommunity: patterns, mechanisms and implications

3

4 Guillermo de Mendoza1,2,*, Riikka Kaivosoja3, Mira Grönroos4, Jan Hjort3, Jari Ilmonen5,

5 Olli-Matti Kärnä3, Lauri Paasivirta6, Laura Tokola3 & Jani Heino7

6

7 1Centre for Advanced Studies of Blanes, Spanish National Research Council (CEAB-CSIC),

8 Accés a la Cala St. Francesc 14, ES-17300 Blanes, Spain.

9 2Laboratoire GEODE, UMR 5602 CNRS, Université Toulouse-Jean Jaurès, 5 allées Antonio

10 Machado, FR-31058 Toulouse, France.

11 3University of Oulu, Geography Research Unit, P.O. Box 3000, FI-90014 Oulu, Finland.

12 4University of Helsinki, Department of Environmental Sciences, Section of Environmental

13 Ecology, Niemenkatu 73, FI-15140 Lahti, Finland.

14 5Metsähallitus, Natural Heritage Services, P.O. Box 94, FI-01301 Vantaa, Finland.

15 6Ruuhikoskenkatu 17, FI-24240 Salo, Finland.

16 7Finnish Environment Institute, Natural Environment Centre, Biodiversity, Paavo Havaksen

17 Tie 3, FI-90570 Oulu, Finland.

18 * Corresponding author: [email protected]

19 20 Summary

21 1. Metacommunity theory focuses on assembly patterns in ecological communities,

22 originally exemplified through four different, yet non-exclusive, perspectives: patch

23 dynamics, species sorting, source-sink dynamics, and neutral theory. More recently, three

24 exclusive components have been proposed to describe a different metacommunity

25 framework: habitat heterogeneity, species equivalence, and dispersal. Here, we aim at

26 evaluating the insect metacommunity of a subarctic stream network under these two

27 different frameworks.

28 2. We first modelled the presence/absence of 47 stream insects in northernmost Finland

29 using binomial generalised linear models (GLMs). The deviance explained by pure local

30 environmental (E), spatial (S), and climatic variables (C) was then analysed across

31 species using beta regression. In this comparative analysis, site occupancy, as well as

32 taxonomic and biological trait vectors obtained from principal coordinate analysis, were

33 used as predictor variables.

34 3. Single-species distributions were better explained by in-stream environmental and spatial

35 factors than by climatic forcing, but in a highly variable fashion. This variability was

36 difficult to relate to the taxonomic relatedness among species or their biological trait

37 similarity. Site occupancy, however, was related to model performance of the binomial

38 GLMs based on spatial effects: as populations are likely to be better connected for

39 common species due to their near ubiquity, spatial factors may also explain better their

40 distributions.

41 4. According to the classical four-perspective framework, the observation of both

42 environmental and spatial effects suggests a role for either mass effects or species sorting

43 constrained by dispersal limitation, or both. Taxonomic and biological traits, including

44 the different dispersal capability of species, were scarcely important, which undermines 45 the patch dynamics perspective, based on differences in dispersal ability between species.

46 The highly variable performance of models makes the reliance on an entirely neutral

47 framework unrealistic as well. According to the three-component framework, our results

48 suggest that the stream insect metacommunity is shaped by the effect of habitat

49 heterogeneity (supporting both species-sorting and mass effects), rather than species

50 equivalence or dispersal limitation.

51 5. While the relative importance of the source-sink dynamics perspective or the species-

52 sorting paradigm cannot be deciphered with the data at our disposal, we can conclude that

53 habitat heterogeneity is an important driver shaping species distributions and insect

54 assemblages in subarctic stream metacommunities. These results exemplify that the use of

55 the three-component metacommunity framework may be more useful than the classical

56 four perspective paradigm in analysing metacommunities. Our findings also provide

57 support for conservation strategies based on the preservation of heterogeneous habitats in

58 a metacommunity context.

59

60 Key-words

61 beta regression, comparative analysis, insects, metacommunity theory, single-species

62 distribution models, stream macroinvertebrates, subarctic streams.

63 64 Introduction

65 Metacommunity theory predicts the assembly of ecological communities according to

66 different perspectives. Originally, this idea was illustrated by Leibold et al. (2004) in the form

67 of four metacommunity perspectives: (1) patch dynamics, which is based on a resource

68 competition-colonisation trade-off among species, thus taking into account species’ dispersal

69 potential (Hanski, 1994); (2) species-sorting along environmental gradients, which relies on

70 differences in environmental tolerance among species (Leibold, 1995); (3) mass effects or

71 source-sink dynamics, whereby species may survive in poor-quality habitats owing to

72 constant immigration from the source populations in high quality habitats (Pulliam 1988);

73 and (4) the neutral theory, where demographic stochasticity solely explains assembly patterns

74 (Hubbell, 2001). Deciphering which of these perspectives is more suitable in the context of

75 metacommunity analysis seems difficult and may well depend on the context of analysis (e.g.

76 spatial extent, biogeographic region, ecosystem type and more; Heino et al., 2015).

77 Nevertheless, the examples of metacommunity perspectives depicted in Leibold et al.

78 (2004) are not mutually exclusive, and represent a fraction of possibilities which can be

79 expanded with the inclusion of species dispersal rates, connectivity, species interactions,

80 disturbance, priority effects, rapid local adaptation, meta-ecosystem dynamics and more

81 (Brown, Sokol, Skelton, & Tornwall, 2017; Logue, Mouquet, Peter, & Hillebrand, 2011). The

82 more recent proposal by Logue et al. (2011) claims that the metacommunity concept is better

83 generalised by three major exclusive components, which decompose the metacommunity

84 framework into (1) environmental heterogeneity, whereby habitat patches differ in

85 environmental attributes; (2) species equivalence, in terms of niche characteristics; and (3)

86 dispersal, referred to as the rate of dispersal among patches. Here, we aim at evaluating

87 species distributions in a subarctic stream insect metacommunity under these two different 88 frameworks (i.e., Leibold et al., 2004 versus Logue et al., 2011), specifically so as to evaluate

89 which of the two is more adequate for the interpretation of our observations.

90 Species distribution models have previously been used to predict community-level

91 properties such as biodiversity (Ferrier & Guisan, 2006). Their accuracy in predicting

92 community-level properties appears to be higher than that of community assembly models,

93 although at a high cost in terms of model complexity (Bonthoux, Baselga, & Balent, 2013,

94 Chapman & Purse, 2011). The accuracy of single-species distribution modelling, however,

95 may also be advantageous to test ecological theories about community assembly mechanisms.

96 This is because accurately modelling the distribution of single species, one at a time, provides

97 the opportunity to proceed with a subsequent comparative analysis across species. Using a

98 comparative analysis, the variation in model performance can be related, for example, to

99 species traits and potential phylogenetic constraints.

100 Stream insect species, in particular, are highly suitable to decipher community

101 assembly processes through the comparative analysis of single-species distribution models

102 (Heino & de Mendoza, 2016). This is because of the high variability among species in

103 tolerance of environmental conditions, as well as resource exploitation, dispersal capability,

104 and habit traits (Merritt & Cummins, 1996; Tachet, Richoux, Bournaud, & Usseglio-Polatera,

105 2010; Schmidt-Kloiber & Hering, 2015; Serra, Cobo, Graça, Dolédec, & Feio, 2016). This

106 variability is valuable in evaluating which community assembly mechanism dominates in

107 each particular context of analysis. Basically, such an analysis might shed light into the

108 relevance of environmental variables, spatial variables, and dispersal capability of species on

109 model performance. Subsequently, this information can be used as an indicator of the

110 preponderance of one community assembly mechanism over another (Figure 1). For example,

111 if many species show similar spatial patterns, and if these species share the same dispersal

112 potential, we can presume that the ability to disperse may be underlying the observed general 113 pattern for these species. This would give us hints about the adequacy to consider one

114 particular metacommunity theory perspective over the others. Within the classical four-

115 perspective framework (Leibold et al., 2004), patch dynamics would likely be suitable in this

116 case, as this perspective relies on the different capability of species to both disperse and

117 exploit resources. Within the metacommunity framework based on three exclusive

118 components (Logue et al., 2011), dispersal would be main driver in this case. Moreover,

119 stream insects are also a diverse group of species, which belong to different insect orders and

120 vary widely in physiological and morphological adaptations (Merritt & Cummins, 1996).

121 Thus, modelling the distribution of single stream insect species and subsequently proceeding

122 with a comparative analysis across species is also a suitable indirect practice to explore

123 possible evolutionary constraints on community assembly processes.

124 In this study, we analysed the distribution of common stream insect species in the

125 metacommunity of a subarctic drainage basin. Species differ widely in their dispersal

126 capability (e.g. passive or active dispersers, aquatic or aerial adults) and tolerance of

127 environmental conditions such as temperature, water flow, or habitat characteristics

128 (Grönroos et al., 2013; Heino, 2005; Heino & Grönroos, 2014). We used environmental,

129 climatic and spatial variables as predictors of the distributions of single stream insect species.

130 Our aim was to elucidate, first, whether or not environmental and spatial factors are relevant

131 for explaining the distribution of stream insect species; and second, whether or not the

132 obtained models can be related to the different dispersal capability, site occupancy (i.e. a

133 gradient of rarity-commonness), and biological and taxonomic traits, of stream insect species.

134 Both considerations were used to evaluate which of the two different metacommunity

135 frameworks, either the one based on four non-exclusive perspectives (Leibold et al., 2004) or

136 the one based on three exclusive axes (Logue et al., 2011), is more adequate to interpret our

137 observations of single species distributions in stream networks (Figure 1). 138

139 Methods

140 Study area

141 The field work for this study (Fig. S1) was conducted in the Tenojoki drainage basin (main

142 stem length: 361 km, basin area: 16377 km2, altitude of sites: from 19 to 285 m a.s.l.) in

143 northernmost Finland (70oN, 27oE). This subarctic drainage basin is close to a natural state,

144 since it is characterised by very small human populations and subsequent little impact from

145 human development. A typical feature of the area are short cool summers and long cold

146 winters (from early November to end of May). The mean annual temperature is about -2oC in

147 the continental areas of the drainage basin, and close to 0oC near the Arctic Ocean (Dankers

148 & Christensen, 2005). Annual precipitation ranges from 310 mm to 410 mm depending on

149 the location in the drainage basin (Mansikkaniemi, 1970). Most of the rainfall and snowmelt

150 enters streams and rivers, as evaporation is generally of minor importance. Vegetation is

151 dominated by mountain birch (Betula pubescens ssp. czerepanowii) woodlands at low altitude

152 and barren tundra at higher altitude, but also peatlands, heathlands and riparian meadows

153 occur commonly. Coniferous pine (Pinus sylvestris) woodlands occur only in scattered

154 locations, mostly in the southern parts of the drainage basin. Wadeable streams and rivers

155 (i.e. channel width < 25 m, water depth < 50 cm) in the area are close to a pristine state,

156 providing excellent possibilities for examining species distributions in natural environmental

157 conditions. We sampled altogether 55 tributary streams for this study (for details, see Kärnä

158 et al., 2015). All these 1st to 5th order tributaries drain into the mainstem of the River

159 Tenojoki or the River Utsjoki, and no site is located in the two mainstem rivers (Fig. S1).

160

161 Field sampling of stream insects 162 We took a 3-minute kick-net sample (net mesh size: 0.3 mm) at each study site (Kärnä et al.,

163 2015) at the same time with the environmental measurements in early and middle of June

164 2012 (see below). The sample for each site consisted of six 30-s subsamples that were

165 divided between main habitats at a riffle site (ca. 50 m2) based on visual inspections of

166 variation in depth, flow, moss cover and particle size. The six subsamples were pooled in the

167 field to obtain a composite sample. Such a sampling method has been shown to be effective

168 in northern streams, allowing to detect patterns in community structure (Heino, Ilmonen, &

169 Paasivirta, 2014) and distributions of single species (Heino & de Mendoza, 2016). Samples

170 were immediately preserved in ethanol in the field and were taken to the laboratory for

171 further processing and identification. Animals were separated from detritus and moss

172 fragments and identified to the lowest possible taxonomic level, mostly species (Kärnä et al.,

173 2015).

174

175 Species considered and species traits

176 We detected 107 insect taxa, of which 87 could be taxonomically determined to species or

177 species group (Kärnä et al., 2015). Insects determined to genus level were discarded as they

178 were considered too likely to include a few species, which is inappropriate to model single-

179 species distributions. Then, we focused on 48 species that occurred at more than 10% of the

180 55 study sites, that is, that occurred in at least six sites. This is because modelling the

181 distribution of species present in less than six sites is likely to produce spurious results and

182 therefore the analysis of these species was considered unreliable (e.g. Pearce & Ferrier,

183 2000). In practice, we could model the occupancies of only 47 species because the mayfly

184 Baetis rhodani occurred at all sites, so we could not use this species to model

185 presence/absence. The 47 stream insect species considered in this study are listed in Table S1. 186 Nomenclature generally follows de Jong et al. (2014) and more specific references for the

187 Simuliidae (Adler & Crosskey, 2016 ; Ilmonen, 2014).

188 Body size class, dispersal potential, functional feeding groups and habit trait groups

189 were considered as species traits (Table S2). Functional feeding groups refer to exploitation

190 of different resources, while habit traits define modes of locomotion and attachment to

191 substrate (Merritt & Cummins, 1996). Body size classes and female dispersal potential

192 followed a previous study (Heino & de Mendoza, 2016), with additional information from

193 Tachet et al. (2010), Schmidt-Kloiber & Hering (2015) and Serra et al. (2016). Female

194 dispersal potential was characterised as being “low” or “high”. In general, all species of the

195 Simuliidae were considered to have high dispersal potential, owing to the fact that their

196 females feed as flying adults, in most cases searching for blood of vertebrates, and hence

197 were assumed here to generally persist much longer as active flyers than the rest of species.

198 In this regard, Baldwin, West, and Gomery (1975) often found their marked Simuliidae

199 females several kilometers away from their natal streams. Owing to their small size, the

200 Simuliidae may also be distributed long distances passively by wind (Crosskey, 1990). All

201 other species were considered as weak dispersers except for the caddisflies Plectrocnemia

202 conspersa and Potamophylax cingulatus, according to the information available for these taxa

203 from previous studies (Gíslason, Hannesdóttir, Munoz, & Pálsson, 2015; Hoffsten, 2004;

204 Müller-Peddinghaus, 2011; Müller-Peddinghaus & Hering, 2013; Schmidt-Kloiber & Hering,

205 2015). Although such information about dispersal abilities of stream insects is rather simple,

206 there is currently no better information available (Schmidt-Kloiber & Hering, 2015; Serra et

207 al., 2016; Tachet et al., 2010). Functional feeding and habit trait groups generally follow

208 Merritt & Cummins (1996).

209

210 Local environmental, climatic, and spatial variables 211 The 55 streams were surveyed during the early northern summer, between early and middle

212 of June in 2012. We measured a set of local (i.e. proximal) environmental variables that have

213 been found important for stream insects in northern drainage basins in previous studies

214 (Heino et al., 2014; Kärnä et al., 2015). These comprised physical habitat and water physico-

215 chemical variables. For physical habitat variables, we measured current velocity (m/s) and

216 depth (cm) at 30 random spots in a riffle site. We also measured mean width of the riffle site

217 based on five cross-channel measurements, evenly spaced across the surveyed riffle site.

218 Bank height and bank slope were measured at the same locations with stream width

219 measurements. Bank height was measured as the height of the lower stream bank, i.e. the

220 height from the water level to the edge of terrestrial vegetation. Bank slope was measured

221 (perpendicular to the stream) as a stream bank rise (cm) over 2 m starting from the edge of

222 terrestrial vegetation. Moss cover (%) and particle size classes (%) were visually estimated at

223 10 squares (1 m2) at random locations in a riffle site. We used a modified Wentworth’s

224 (1922) scale of particle size classes: sand (0.25–2 mm), gravel (2–16 mm), pebble (16–64

225 mm), cobble (64–256 mm) and boulder (256–1,024 mm). Based on the visual estimates for

226 each square, we calculated mean values for each particle size class and moss cover at a site

227 and used these mean values in species distribution modelling. We also visually estimated

228 shading (%) by riparian vegetation and proportion of riparian deciduous trees (%). For

229 physico-chemical properties, we measured pH, conductivity and water temperature at each

230 site in the field using a YSI device model 556 MPS (YSI Inc., Ohio, USA) and took

231 additional water samples during the field campaign for further analysis. Water samples were

232 frozen at the end of the day at the Kevo Field Station situated in the northern part of the study

233 area, and were later analysed for total nitrogen, colour, iron and manganese in the laboratory

234 of the Finnish Environment Institute in Oulu following Finnish national standards (National

235 Board of Waters, 1981). 236 We also included three climatic variables, including annual air temperature sum above

237 5oC (growing degree days), mean annual air temperature and mean July air temperature for

238 the period 1981–2010. These variables were calculated in ArcMap 10.2 for each site from a

239 gridded (1 x 1 km) climate data provided by the Finnish Meteorological Institute (Pirinen et

240 al., 2012). The gridded climate data were produced using meteorological station observations

241 and Kriging interpolation (e.g. Aalto, Pirinen, Heikkinen, & Venäläinen, 2013). The selected

242 climatic variables are likely to be important for the distributions of insects in this subarctic

243 area, where temperature is closely associated with insect life cycles (Danks, 2007).

244 Spatial variables were distance-based Moran’s Eigenvector Maps (db-MEM) based on

245 geographical distances among sites (Dray, Legendre & Peres-Neto, 2006). These spatial db-

246 MEM variables were obtained with the function “PCNM” of the R package “PCNM”

247 (Legendre, Borcard, Blanchet, & Dray, 2013; R Core Team, 2013). We used the largest

248 distance in the minimum spanning tree, keeping all sites connected, as the truncation

249 threshold. Spatial db-MEM variables represent structures of autocorrelation at all spatial

250 scales. Only those spatial db-MEM variables showing significant positive autocorrelation

251 were included in subsequent modelling (Borcard, Gillet & Legendre, 2011), resulting in 13

252 spatial variables (Figure 2). Based on eigenvalues and bubble plot maps, the spatial variables

253 can be divided into those ranging from large-scale spatial structures (e.g. V1, V2) and those

254 showing very small scale spatial patterns (e.g. V12, V13).

255 Prior to modelling species distribution, we eliminated strongly correlated (i.e. Pearson

256 r > .7) predictor variables from the sets of local environmental and climatic variables (see

257 Dormann et al., 2013). Hence, we removed one variable (i.e. annual temperature sum) from

258 the climatic variables and four variables (i.e. water iron, colour, conductivity and boulders)

259 from the stream environmental variables. The spatial variables were already not mutually

260 correlated (Borcard et al., 2011). 261

262 Modelling species distributions

263 The distribution (i.e. presence/absence) of each species was modelled using binomial

264 generalised linear models (i.e. binomial GLMs with logit link function), using separately

265 local environmental, climatic and spatial variables, with the R package “Rcmdr” (Fox, 2005).

266 The deviance explained for each species was thus obtained for each binomial GLM with each

267 of these three different subsets of variables (Figure 2). The variables selected for each

268 species’ model were based on forward selection and Bayesian Information Criterion (BIC),

269 separately for each variable group (i.e. environmental, climate and spatial). BIC values were

270 used because they prevented the selection of too complex models in our case, in contrast to

271 AIC (results not shown), which is often the case under large sample sizes (Burnham &

272 Anderson, 2004). Moreover, the target model under BIC selection does not depend on sample

273 size, in contrast to AIC (Burnham & Anderson, 2004). Therefore, AIC may be problematic in

274 our case as we aim at comparing model performance between species, which may differ in

275 the number of presences and absences. Also, deviating observations were removed from

276 some species’ models if they had Cook’s distance values > 1 and hence affected profoundly a

277 few models (Cook, 1977). For environmental variables, we registered whether the effect was

278 positive or negative on species distributions. We then used the selected variables of these

279 three subsets (i.e. local environmental, climatic and spatial) to perform variation (deviance)

280 partitioning by subtraction, similarly as performed in multivariate contexts (Legendre &

281 Legendre, 2012). Specifically, the deviance accounted for subset A, subset B, and subset A

282 and B together, was computed, so as to obtain the different fractions of variation solely

283 explained by each subset (i.e. unshared with other subsets). We eventually obtained adjusted

284 D2 values (Guisan & Zimmermann, 2000; Legendre & Legendre, 2012) which could be

285 attributed to pure local environmental (E), climatic (C) or spatial effects (S), as well as to 286 total effects combining the three subsets of pure effects and their joint effects (E+C+S

287 effects; Figure 2). Modelling methods other than GLMs could have been possible, yet species

288 probably show linear responses to the environmental predictors due to the fact that they are

289 on the edge of their geographical and ecological distributions, making GLMs adequate.

290 Adding quadratic terms to binomial models is unlikely to change results substantially in these

291 situations (e.g. Pulido, Riera, Ballesteros, Chappius, & Gacia, 2015), and increase the

292 difficulty of interpretation of the results. Also, deviance partitioning is easy to accomplish

293 when this is based on GLMs.

294

295 Comparative analysis across species

296 We performed a comparative analysis across species using beta regression (Ferrari & Cribari-

297 Neto, 2004), where the adjusted D2 values obtained with previous binomial GLMs were used

298 as the dependent variable to be explained by site occupancy, taxonomic vectors or species

299 trait vectors (Figure 2). These vectors were obtained separately from Principal Coordinate

300 Analysis (PCO). Using the taxonomic relatedness of species, a taxonomic relatedness matrix

301 was built using the function “taxa2dist” in the R package “vegan” (Oksanen et al., 2013), and

302 taxonomic vectors were handled as continuous PCO vectors with the function “pco” in the R

303 package “ecodist” (Goslee & Urban, 2007). The first four taxonomic eigenvectors were

304 selected as these had much higher eigenvalues than the rest (Fig. S2). Similarly, species trait

305 vectors were also computed using body size class, dispersal potential, functional feeding

306 groups, and habit trait groups (Table S2). Species traits were considered as regular factors,

307 except body size class which was considered as an ordered factor, to obtain a distance matrix

308 based on Gower’s metric with the function “daisy” of the R package “cluster” (Maechler,

309 Rousseeuw, Struyf, Hubert, & Hornik, 2013), and eventually trait PCO vectors with the

310 function “cmdscale”. The four trait eigenvectors obtained were considered for further 311 statistical analyses. The variation in adjusted D2 values across species that could be attributed

312 to pure E, pure C, pure S, or E+C+S effects was fitted on site occupancy, the four taxonomic

313 and four species trait vectors selected, using beta regression with the function “betareg” of

314 the R package “betareg” (Cribari-Neto & Zeileis, 2010). Beta regression is adequate when the

315 response variable (in this case, the adjusted D2 values) is constrained between 0 and 1.

316 We compared the explained variation by pure E, C, and S effects with a Kruskal-

317 Wallis test, with additional Mann-Whitney tests for subsequent pair-wise comparisons

318 between groups. Non-parametric tests were chosen since adjusted D2 values data departed

319 from normality following the Shapiro-Wilk test (Zar, 1984). We also analysed the univariate

320 relationships between site occupancy, body size, dispersal potential, broad taxonomic insect

321 groups, functional feeding groups, habit trait groups, and taxonomic and trait vectors.

322 Depending on the continuous (e.g. site occupancy) or categorical (e.g. habit trait group)

323 nature of the variables involved, we followed Kruskal-Wallis tests, Mann-Whitney tests,

324 Fisher’s exact test or Spearman correlations, as these variables were generally not normally

325 distributed (Zar, 1984).

326

327 Results

328 Single species models

329 Local environmental and spatial effects accounted for a higher variation in species

330 distributions (16.1% and 12.6% in average, respectively) than did climatic effects (5.4%; p <

331 .001, Kruskal-Wallis test), whereas the average deviance explained did not differ

332 significantly between local environmental and spatial effects (p = .125, Mann-Whitney test)

333 (Table S3). The local environmental factors most frequently selected in explaining species

334 distributions were water temperature, shading, and to a lesser extent, stream width, cobbles

335 and moss (Figure 3). The spatial variables most often selected were better represented by 336 large-scale spatial variables within the Tenojoki drainage basin (e.g. V1, V2) than by small-

337 spatial scale variables (e.g. V12, V13), as also shown in Figure 3. Amongst the climate

338 variables, mean annual temperature was significant in explaining the distribution of 32

339 species, and July air temperature of 20 species (not shown in Fig. 3).

340 The adjusted deviance explained by binomial GLMs was highly variable across

341 species and difficult to relate to particular taxonomic groups (Table S3). For example, local

342 environmental effects were particularly relevant for the stonefly Siphonoperla burmeisteri

343 (i.e. accounting for 66.7% of adjusted D2 values), the mayfly Heptagenia dalecarlica

344 (50.2%), and the blackfly Prosimulium hirtipes (37.7%), whereas spatial effects were most

345 relevant for the caddisfly Rhyacophila nubila (41.2%), the stonefly Brachyptera risi (31.5%),

346 and the chironomid midge Cardiocladius capucinus (28.6%). Climate effects were also

347 highly variable. They were generally low (see above), and accounted for more than 20% of

348 adjusted D2 values in only three cases: the stoneflies Diura nanseni and Siphonoperla

349 burmeisteri (34.4% and 20.1%, respectively), and the chironomid midge Orthocladius

350 rivicola (28.9%). Combining all effects, binomial GLMs explained on average 37.8% of the

351 null deviance (Table S3).

352

353 Comparative analysis across species models

354 The highly variable species-local environment and species-climate relationships in binomial

355 GLMs were not accounted for by site occupancy, or by taxonomic and trait vectors, in the

356 beta regression analysis (Table 1). The deviance explained by spatial variables was, however,

357 significantly (i.e. p < .05) accounted for by site occupancy (Table 1). The influence of TAX-

358 PCO4 and TRA-PCO2 on the adjusted D2 values predicted by spatial effects in binomial

359 GLMs was significant as well. Also, the influence of TAX-PCO3 was marginally significant

360 (i.e. p < .10), remaining like this in the binomial GLMs based on all variables combined 361 (Table 1). However, when repeating the beta regression analysis by using only the significant

362 variables selected (i.e. site occupancy, TAX-PCO3, TAX-PCO4, and TRA-PCO2), only site

363 occupancy was statistically significant (p = .017), but not TAX-PCO3, TAX-PCO4 or TRA-

364 PCO2 (p = .943, p = .175, and p = .449, respectively, results not shown in Table 1).

365 Analysing through beta regression the univariate relationship of these variables with the

366 adjusted D2 values of binomial GLMs based on spatial effects produced a similar result (site

367 occupancy, p = .036, Fig. S3; TAX-PCO3, TAX-PCO4, and TRA-PCO2, p = .760, p = .660,

368 and p = .524, respectively, results not shown). This univariate relationship between the

369 adjusted D2 values and site occupancy was not observed when the adjusted D2 values of

370 binomial GLMs were referred to environmental or climate effects (Fig. S3). No statistical

371 significance was observed either for univariate relationships between separate species traits

372 and the adjusted D2 values in binomial GLMs, with the sole exception of body size (Fig. S3).

373 The TAX-PCO3 vector showed the highest species scores for blackflies (Simuliidae)

374 and the lowest for mayflies (Ephemeroptera), and was strongly correlated (p < .001) to

375 dispersal potential (Table S5, Figure 4a). In contrast to this taxonomic vector, TAX-PCO4

376 showed the highest species scores for both blackflies and mayflies (Figure 4b), and was

377 strongly correlated to site occupancy (p = .007, Table S5). Finally, TRA-PCO2 reflects the

378 influence of functional feeding groups and body size on model performance (Figure 4c), as

379 indicated by the strong correlation of both variables (i.e. p < .001) with this trait vector (Table

380 S5).

381

382 Discussion

383 Single species models

384 Our results indicated that single species distributions of stream insects are highly variable in

385 terms of predictability, as well as the significant environmental and spatial predictors 386 underlying such distributions. There was no evident association between model accuracy and

387 particular taxonomic groups (Table S3). Nevertheless, a few generalisations can be

388 highlighted with regard to the results obtained. For example, water temperature and shading,

389 and to a lesser extent, stream width, cobbles and moss, were more relevant as environmental

390 predictors of species distributions than stream flow or water chemistry variables (Table S3,

391 Figure 3). This is in line with the well-known influence of temperature and resource

392 availability on insect life cycles at high latitudes (Danks, 2007) and indicates the influence of

393 species sorting processes along these environmental gradients. Resource availability is

394 represented in our case by shading, which indicates the proximity of terrestrial vegetation and

395 hence is a surrogate of availability of allochthonous resources from terrestrial origin for

396 aquatic insect larvae. This typically corresponds with a situation of a low-order stream which,

397 as in our case, is influenced strongly by terrestrial material from riparian vegetation which is

398 then taken as food resource by shredders, hence promoting their dominance (Vannote,

399 Minshall, Cummins, Sedell, & Cushing, 1980). Shading may also be inversely related to

400 primary productivity, but in this study, we found that the relationship of species distribution

401 with shading was always positive (Figure 3), suggesting that rather than biofilm production, it

402 is the external input of terrestrial material from riparian birch tree abundance what is likely

403 driving species distributions. In our case, shading was selected as a significant variable in

404 binomial models for some predators (Isoperla difformis and Plectrocnemia conspersa) and

405 shredders (Leuctra spp.), for some collector-gatherers (Corynoneura lobata-type,

406 Eukiefferiella devonica-group, Orthocladius rhyacobius-group and Tvetenia discoloripes),

407 and for some collector-filterers (Philopotamus montanus and Prosimulium hirtipes) (Table

408 S3). These latter groups perhaps benefit indirectly from the increase in potential resources

409 that the variable “shading” represents for shredders, for example, through the enhancement of 410 nutrient re-cycling by shredding coarse plant litter (Covich, Palmer, & Crowl, 1999; Wallace

411 & Webster, 1996).

412 Spatial variables were also relevant for the distributions of some species. Specifically,

413 large-scale spatial variables were more important than small-scale variables in explaining

414 species distributions in our study (Table S3, Figure 3). At a larger spatial extent (ca. 500 km

415 latitudinal gradient), previous findings indicate a stronger relevance of environmental factors,

416 compared to spatial restrictions, on single-species distributions (Heino & de Mendoza, 2016).

417 This is perhaps not surprising because increasing the spatial extent may have a strong positive

418 effect on the relevance of niche processes through larger environmental gradients (Chase,

419 2014). However, increasing the spatial extent may also preclude species to reach

420 environmentally suitable locations owing to dispersal limitation, and thus the relative

421 contribution of both environmental and spatial constraints on species distributions does not

422 always vary predictably with spatial scale (Alahuhta & Heino, 2013).

423

424 Comparative analysis across species

425 Comparative analysis across the species models showed a clear relationship between model

426 performance and site occupancy. Specifically, the binomial GLMs that we built upon spatial

427 variables could be related to site occupancy, and to a lesser extent, to taxonomic and trait

428 vectors, whereas none of these variables was significantly related to model performance

429 when models were based on local environmental or climate variables (Table 1). At first

430 glance, our results also suggested both a slight influence of female dispersal potential (related

431 to the taxonomic vector TAX-PCO3), and a potential influence of functional feeding groups

432 and body size (related to the trait vector TRA-PCO2), on the performance of models based on

433 spatial variables. The taxonomic vector TAX-PCO3 perhaps relates to female dispersal

434 potential, as species scores along this vector were much higher for the blackflies than for the 435 rest of species, and lowest for the mayflies (Figure 4). Blackflies are possibly the best active

436 dispersers among all the insects we considered, because females feed as flying adults and in

437 most species they must actively search for blood meals, often several kilometers away from

438 their natal streams (Baldwin et al., 1975). However, adult mayflies, do not feed and often

439 have extremely short life spans (Brittain, 1990). Therefore, it seems reasonable to assume that

440 blackflies may actively disperse better than mayflies. Site occupancy and dispersal potential

441 were not correlated (Table S5), and both taxa were the ones with highest number of sites

442 occupied (Fig. S4). In contrast, mayflies differed in site occupancy from non-biting midges

443 (Chironomidae) (Fig. S4), despite species in both groups can be considered weak active

444 dispersers, as chironomid adults are also short-lived and generally weak active fliers

445 (Armitage, 1995). On the other hand, the trait vector TRA-PCO2 suggests an influence of

446 feeding behaviour and body size (Figure 4, Table S5) on model performance. This is because

447 the exploitation of food resource from terrestrial origin (i.e. shredders) would facilitate the

448 development of more complex trophic food webs with the inclusion of predators (Figure 4).

449 This would also contribute to the positive association of body size to TRA-PCO2 (Fig. 4), as

450 the largest insects we found are either predators or shredders (Table S2).

451 Nevertheless, it is important to note that taxonomic and trait vectors had a

452 comparatively much weaker effect on predictability by spatial variables than that of site

453 occupancy. In fact, not only did site occupancy attain a higher statistical significance (Table

454 1), but it could also be partly related to the capability of the taxonomic vector TAX-PCO4 to

455 account for the adjusted D2 values of binomial GLMs because these two predictor variables

456 were significantly correlated (Table S5). Moreover, when repeating the beta regression

457 analysis by using only the significant variables selected (i.e. site occupancy, TAX-PCO3,

458 TAX-PCO4, and TRA-PCO2), only site occupancy was statistically significant, indicating

459 that the influence of taxonomic and trait vectors on model performance is rather weak. 460 Analysing through beta regression the univariate relationship of these variables with the

461 adjusted D2 values of binomial GLMs based on spatial effects again resulted in site

462 occupancy as the only significant variable (see Results above). Therefore, we must conclude

463 that any potential effect of taxonomic and trait vectors on model performance, including the

464 effect of female dispersal potential and body size, and that of functional feeding groups, must

465 be considered with caution: their statistical significance only appears after controlling for site

466 occupancy and the other variables considered in the full model of beta regression. In this

467 regard, the fact that Baetis rhodani is a widespread mayfly, which could not be modelled

468 because it was present at all sites, also gives support to the idea that dispersal abilities are not

469 so important in structuring invertebrate assemblages in high-latitude drainage basins. This is

470 because it demonstrates that mayfly species can be widespread, despite being rather weak

471 active dispersers. We also acknowledge that the rarest species (i.e. present in less than six

472 sites) were not modelled because models based on such small number of presences were

473 considered unreliable (e.g. Pierce & Ferrier, 2000). However, excluding these species does

474 not undermine the conclusion that the distributions of most common species are better

475 accounted for by models based on spatial variables than that of not-so-common species. In

476 fact, we effectively modelled 47 out of the 86 taxa available at the species (most cases) or

477 species-group (few cases) taxonomic resolution, comprising 55% of cases, which is a

478 representative subset of species in the entire metacommunity.

479

480 Approaching the suitability of metacommunity analysis frameworks

481 With the information above about single-species distribution models and subsequent

482 comparative analysis across species, it is possible to proceed with the evaluation of the

483 suitability of the two different frameworks of metacommunity analysis (Figure 1) considered

484 here: (1) the classical approach exemplified by the four different non-exclusive perspectives 485 described by Leibold et al. (2004) or (2) the three exclusive components as proposed by

486 Logue et al. (2011).

487 Among the four different metacommunity perspectives of the Leibold et al. (2004)

488 framework, neutral theory and patch dynamics do not rely on the effect of environmental

489 variables, in contrast to species sorting and source-sink dynamics, the latter of which also

490 incorporating a strong influence of spatial effects (Figure 1a). In our study, single-species

491 models often relied on the effect of environmental variables, particularly temperature and

492 shading, while being also dependent on large-scale spatial variables (Figure 3). As

493 environmental and spatial factors are both relevant for the distribution of species, this result

494 suggests that either species sorting along spatially structured environmental gradients, or

495 source-sink dynamics between populations of high-quality and low-quality habitats, are both

496 likely as important processes driving metacommunities. Then, the comparative analysis

497 across species showed that site occupancy is responsible for the observed differences in the

498 relevance of spatial variables on species distributions (Table 1). This suggests that common

499 species would be better able than rare species to maintain populations in low-quality habitats

500 through constant immigration, favouring the source-sink dynamics perspective over species

501 sorting.

502 Although species-sorting processes cannot be completely discarded because of the

503 demonstrated influence of environmental variables in many cases, deviance partitioning

504 suggests that the pure effects of environmental and spatial factors on species distributions are

505 stronger than their joint effects (Table S3). Also, the effect of spatial variables was better

506 explained than that of environmental factors by our explanatory variables, particularly site

507 occupancy, in the comparative analysis. These results slightly undermine the idea of species-

508 sorting across spatially structured environmental gradients as the most important process

509 shaping metacommunities. In any case, the neutral theory, which relies entirely on spatial 510 dynamics, is unlikely. As the dispersal potential of species has a rather weak effect on model

511 accuracy, patch dynamics can be discarded as well as a suitable perspective of

512 metacommunity analysis in our case. It should be acknowledged, however, that the difficulty

513 to explain model performance with dispersal ability can also be a consequence of the

514 coarseness of the dispersal measures currently available for freshwater invertebrates

515 (Schmidt-Kloiber & Hering, 2015; Serra et al., 2016; Tachet et al., 2010). Moreover, the

516 different metacommunity paradigms from Leibold et al. (2004) may always act

517 simultaneously to a certain extent along a continuum (Figure 1a) rather than being distinct

518 and mutually exclusive options (Brown, Sokol, et al., 2017; Gravel, Canham, Beaudet, &

519 Messier, 2006; Logue et al., 2011).

520 Spatial autocorrelation may appear not only as a consequence of mass effects or

521 species sorting along spatially structured environmental gradients when the spatial scale is

522 not very large, but also as a consequence of dispersal limitation at very large spatial scales

523 (Heino et al., 2015). Nevertheless, some insect species found in this study exemplify well the

524 potential importance of the source-sink dynamics for metacommunities in subarctic streams,

525 independently of their dispersal capability. For example, six blackfly species were examined

526 (Table S1), of which five were present in more than 50% of sites, three of them in 75% of

527 sites or more (Table S2). Thus, blackfly species in subarctic streams have successfully spread

528 widely, which is advantageous to maintain metapopulations through source-sink dynamics.

529 On the other hand, the mayflies are as widespread as the blackflies (Fig. S4), but far less

530 capable of active dispersal. This suggests that the dispersal capability of species does not

531 determine the metapopulation dynamics, whereas site occupancy probably does so. Spatial

532 autocorrelation patterns have been described for the blackflies at small spatial scales, driven

533 by strong effects of inter-specific competition for oviposition sites, and subsequent priority

534 effects at the community level (McCreadie & Adler, 2012). The importance of priority effects 535 for the blackflies reinforces the idea of the relevance of site occupancy for community

536 dynamics, where rare species are in clear disadvantage for habitat recolonisation.

537 Alternative to the framework of Leibold et al. (2004), we can interpret our results

538 under the framework of Logue et al. (2011), whereby three different and mutually exclusive

539 components can be used to analyse metacommunities: species equivalence, habitat

540 heterogeneity and dispersal (Figure 1b). In our case, this alternative framework makes

541 interpretation of the results much easier. At the very least, we can conclude that species

542 equivalence is unlikely to play any role in metacommunity dynamics, similarly to discarding

543 neutral theory under the Leibold et al. (2004) framework. Dispersal can also be discarded, yet

544 again with caution due to the current lack of high resolution dispersal measures for freshwater

545 invertebrates (Schmidt-Kloiber & Hering, 2015; Serra et al., 2016; Tachet et al., 2010). Thus,

546 the main difference in the interpretation of the results with this alternative framework is that

547 we can now be certain about the role of habitat heterogeneity, while under the Leibold et al.

548 (2004) framework it is more difficult to discern whether species sorting or source-sink

549 dynamics is the dominant process. Habitat heterogeneity is indeed related to both

550 mechanisms. In fact, using habitat heterogeneity in space and time as the templet for

551 ecological strategies (Southwood, 1977) could be the framework of choice in situations

552 where it is difficult to discern species sorting processes from source-sink dynamics.

553

554 Alternative approaches, caveats and conclusions

555 Emergent properties at the community level are difficult to discern from field observational

556 data alone. In this regard, population genetics can be very useful in order to gain confidence

557 about the distinction between, e.g., source-sink dynamics and species sorting processes. This

558 is because population genetic studies could be used to estimate the relative contribution of

559 immigrants from nearby populations to the genetic variability of the population under study 560 (Bunn & Hughes, 1997; Hughes, Huey, & Schmidt, 2013; Hughes, Schmidt, & Finn, 2009).

561 Genetic analyses would probably provide the opportunity for a more robust interpretation of

562 our results. Genetic studies, however, are difficult to accomplish with stream insects in the

563 field when the idea is to compare many species at a time, and they are far more expensive

564 than the comparative approach of single species distributions we considered here. Therefore,

565 the comparative approach presented here can be used as a first step to explore the relative

566 contribution of environmental and spatial factors on species distributions, without using

567 expensive and time-consuming genetic analyses. In fact, by using the comparative approach

568 we can certainly conclude that the dispersal capability of species and neutral theory play little

569 role in shaping subarctic stream insect metacommunities. Rather, it is habitat heterogeneity,

570 which influences mass effects and/or species sorting processes, that matters. Subsequently,

571 the results of our study strongly recommend the preservation of habitat heterogeneity as the

572 conservation strategy to maintain biodiversity in these ecosystems.

573 Nevertheless, it should be acknowledged that one shortcoming of single-species

574 distribution modelling is that it does not consider the influence of species interactions in

575 structuring ecological communities. Stream ecology has considered that severe environmental

576 conditions may weaken the potential effects of biotic interactions in structuring communities

577 (Peckarsky, 1983). However, more recent findings pose doubts as to whether this is actually

578 true (Cadotte & Tucker, 2017; Thomson et al., 2002). In fact, biotic interactions can

579 reproduce patterns of community structure essentially identical to what it could be expected

580 from environmental filtering alone. This is because environmental changes may affect

581 population growth rates of competing species in opposite ways, and this may cause the

582 exclusion of some species that would otherwise be able to coexist (Cadotte & Tucker, 2017).

583 There exists also evidence indicating that biotic interactions limit the geographical range

584 expansion of species facing environmental changes (Pigot & Tobias, 2013; Sexton et al., 585 2009). Overall, this suggests that inter-specific interactions may also play a role in our case,

586 although the abundances of insect larvae in subarctic streams are typically low (see also

587 Heino & Grönroos, 2017) and may thus result in weak density-dependent interactions among

588 species (see also Morin, 2011).

589 Our study considered tributary streams draining into two linear sub-elements of a

590 larger river network (Fig. S1). However, there exists growing concern about the potential role

591 of the entire dendritic river networks in shaping biodiversity patterns, community structure

592 and species distributions (Altermatt, 2013; Altermatt & Fronhofer, 2017; Brown, Wahl, &

593 Swan, 2017; Jamoneau, Passy, Soininen, Lebuocher, & Tison-Rosebery, 2017; Schmera et

594 al., 2017). For example, the consideration of whole river networks may unveil a more

595 preeminent role for spatial factors in community assembly, undermining the role of

596 environmental filtering. Therefore, studies conducted across whole dendritic networks could

597 be more in line with neutral theory, as shown by Muneepeerakul et al. (2008) for fish

598 communities, yet no environmental variable was truly considered in that study. Although we

599 focused on tributary streams draining into the main river, the consideration of whole dendritic

600 networks may help us to perceive more accurately the real connectivity pathways between

601 isolated patches. This connectivity may have consequences for metacommunity stability with

602 respect to a situation where only a linear component of this network is acting (Fagan, 2002).

603 Also, dispersal along dendritic networks implies more variability in local richness with strong

604 consequences also for community differentiation among patches (Carrara, Altermatt,

605 Rodriguez-Iturbe, & Rinaldo, 2012; Seymour, Fronhofer, & Altermatt, 2015). In any case,

606 dispersal processes in stream networks may depend on the organism group considered

607 (Schmera et al. 2017).

608 In the case of stream insects, the taxa considered and the taxonomic resolution

609 achieved prior to species-distribution modelling, may also have important consequences on 610 our perception of the influence of dendritic riverine networks on biodiversity patterns (Kaelin

611 & Altermatt, 2016). Here, some taxa were discarded as it was not possible to determine the

612 species. Provided that the influences of dendritic landscapes and biotic interactions (discussed

613 above) may strongly affect how we understand the reality of community assemblages, it is

614 essential to use the best taxonomic resolution possible to make accurate inferences about the

615 mechanisms truly governing the observed patterns. In fact, the criterion of ‘best taxonomic

616 resolution possible’ used in our modelling endeavours is a fundamental requirement to draw

617 robust conclusions to be applied in biodiversity conservation.

618 Finally, for biodiversity conservation, it is essential to focus on maintaining habitat

619 heterogeneity because it appears to determine metacommunity organization (Kärnä et al.

620 2015) and species distributions (Heino & de Mendoza, 2016) in streams at high latitudes.

621 Unless habitat heterogeneity is not considered (along with potentially important effects of

622 dendritic network structure and biotic interactions), conservation plans may fall short and not

623 result in desired outcomes. .

624

625 Acknowledgements

626 We thank Sirkku Lehtinen and Marja Lindholm for help in the field or the laboratory. We

627 also thank two anonymous reviewers and the Guest Editor Florian Altermatt for their very

628 helpful and constructive comments on an earlier version of this manuscript. Kevo Subarctic

629 Research Station provided facilities during the field work. This study is part of the project

630 “Spatial scaling, metacommunity structure and patterns in stream communities” that was

631 supported financially by a grant from the Academy of Finland. Further support was provided

632 by grants (no: 273557, no: 267995 and no: 285040) from the Academy of Finland. The

633 authors declare no conflict of interest.

634 635 References

636 Aalto, J., Pirinen, P., Heikkinen, J., & Venäläinen, A. (2013) Spatial interpolation of monthly

637 climate data for Finland: comparing the performance of kriging and generalized

638 additive models. Theoretical and Applied Climatology, 112, 99–111.

639 Adler, P.H., & Crosskey, R.W. (2016) World blackflies (Diptera: Simuliidae): A

640 comprehensive revision of the taxonomic and geographical inventory. Retrieved from

641 http://www.clemson.edu/cafls/biomia/pdfs/blackflyinventory.pdf.

642 Alahuhta, J., & Heino, J. (2013) Spatial extent, regional specificity and metacommunity

643 structuring in lake macrophytes. Journal of Biogeography, 40, 1572–1582.

644 Altermatt, F. (2013) Diversity in riverine metacommunities: a network perspective. Aquatic

645 Ecology, 47, 365–377.

646 Altermatt, F., & Fronhofer, E. A. (2017) Dispersal in dendritic networks: Ecological

647 consequences on the spatial distribution of population densities. Freshwater Biology,

648 63, 22–32. https://doi.org/10.1111/fwb.12951

649 Armitage, P. D. (1995) Behaviour and ecology of adults. In P. D. Armitage, P. S. Cranston, &

650 L. C. V. Pinder (Eds.), The Chironomidae: Biology and Ecology of Non-Biting

651 Midges (pp. 194–224). London: Chapman & Hall.

652 Baldwin, W. F., West, A. S., & Gomery, J. (1975) Dispersal pattern of black flies (Diptera:

653 Simuliidae) tagged with 32P. The Canadian Entomologist, 107, 113–118.

654 Bonthoux, S., Baselga, A., & Balent, G. (2013) Assessing community-level and single-

655 species models predictions of species distributions and assemblage composition after

656 25 years of land cover change. PLoS ONE, 8, e54179.

657 Borcard, D., Gillet, F., & Legendre, P. (2011) Numerical Ecology with R. New York:

658 Springer. 659 Brittain, J. E. (1990) Life history strategies in Ephemeroptera and Plecoptera. In I. C.

660 Campbell (Ed.), Mayflies and Stoneflies (pp. 1–12). Dordrecht: Kluwer Academic

661 Publishers.

662 Brown, B. L., Sokol, E. R., Skelton, J., & Tornwall, B. (2017) Making sense of

663 metacommunities: dispelling the mythology of a metacommunity typology.

664 Oecologia, 183, 643–652.

665 Brown, B. L., Wahl, C., & Swan, C. M. (2017). Experimentally disentangling the influence

666 of dispersal and habitat filtering on benthic invertebrate community structure.

667 Freshwater Biology, 63, 48–61. https://doi.org/10.1111/fwb.12995

668 Bunn, S. E., & Hughes, J. M. (1997) Dispersal and recruitment in streams: evidence from

669 genetic studies. Journal of the North American Benthological Society, 16, 338–346.

670 Burnham, K. P., & Anderson, D. R. (2004) Multimodel inference: understanding AIC and

671 BIC in model selection. Sociological Methods & Research, 33, 261–304.

672 Cadotte, M. W., & Tucker, C. M. (2017) Should environmental filtering be abandoned?

673 Trends in Ecology & Evolution, 32, 429–437.

674 Carrara, F., Altermatt, F., Rodriguez-Iturbe, I., & Rinaldo, A. (2012) Dendritic connectivity

675 controls biodiversity patterns in experimental metacommunities. Proceedings of the

676 National Academy of Sciences of the United States of America, 109, 5761–5766.

677 Chapman, D. S., & Purse, B. V. (2011) Community versus single-species distribution models

678 for British plants. Journal of Biogeography, 38, 1524–1535.

679 Chase, J. M. (2014) Spatial scale resolves the niche versus neutral theory debate. Journal of

680 Vegetation Science, 25, 319–322.

681 Cook, R. D. (1977) Detection of influential observation in linear regression. Technometrics,

682 19, 15–18. 683 Covich, A. P., Palmer, M. A., & Crowl, T. A. (1999) The role of benthic invertebrate species

684 in freshwater ecosystems: Zoobenthic species influence energy flows and nutrient

685 cycling. BioScience, 49, 119-127.

686 Cribari-Neto, F., & Zeileis, A. (2010) Beta regression in R. Journal of Statistical Software,

687 34, 1–24.

688 Crosskey, R. W. (1990) The Natural History of Blackflies. John Wiley & Sons, Chichester.

689 Dankers, R., & Christensen, O.B. (2005) Climate change impacts on snow coverage,

690 evaporation and river discharge in the sub-arctic Tana basin, Northern Fennoscandia.

691 Climatic Change, 69, 367–392.

692 Danks, H. V. (2007) How aquatic insects live in cold climates. The Canadian Entomologist,

693 139, 443–471.

694 de Jong, Y., Verbeek, M., Michelsen, V., Bjørn, P. P., Los, W., Steeman, F., Bailly, N.,

695 Basire, C., Chylarecki, P., Stloukal, E., Hagedorn, G., Wetzel, F. T., Glöckler, F.,

696 Kroupa, A., Korb, G., Hoffmann, A., Häuser, C., Kohlbecker, A., Müller, A.,

697 Güntsch, A., Stoev, P., & Penev, L. (2014) Fauna Europaea: All animal species on the

698 web. Biodiversity Data Journal, 2, e4034.

699 Dormann, C. F., Elith, J., Bacher, S., Buchmann, C., Carl, G., Carré, G., García Marquéz, J.

700 R., Gruber, B., Lafourcade, B., Leitão, P. J., Münkemüller, T., McClean, C., Osborne,

701 P. E., Reineking, B., Schröder, B., Skidmore, A. K., Zurell, D., & Lautenbach, S.

702 (2013) Collinearity: A review of methods to deal with it and a simulation study

703 evaluating their performance. Ecography, 36, 27–46.

704 Dray, S., Legendre, P., & Peres-Neto P. R. (2006) Spatial modelling: A comprehensive

705 framework for principal coordinate analysis of neighbour matrices (PCNM).

706 Ecological Modelling, 196, 483–493. 707 Fagan, W. F. (2002) Connectivity, fragmentation, and extinction risk in dendritic

708 metapopulations. Ecology, 83, .3243–3249.

709 Ferrari, S. L. P., & Cribari-Neto, F. (2004). Beta regression for modelling rates and

710 proportions. Journal of Applied Statistics, 31, 799–815.

711 Ferrier, S., & Guisan, A. (2006) Spatial modelling of biodiversity at the community level.

712 Journal of Applied Ecology, 43, 393–404.

713 Fox, J. (2005) The R Commander: A Basic Statistics Graphical User Interface to R. Journal

714 of Statistical Software, 14(9), 1–42.

715 Gíslason, G. M., Hannesdóttir, E. R., Munoz, S. S., & Pálsson, S. (2015) Origin and dispersal

716 of Potamophylax cingulatus (Trichoptera: Limnephilidae) in Iceland. Freshwater

717 Biology, 60, 387–394.

718 Goslee, S.C. & Urban, D.L. (2007) The ecodist package for dissimilarity-based analysis of

719 ecological data. Journal of Statistical Software, 22(7), 1–19.

720 Gravel, D., Canham, C. D., Beaudet, M., & Messier, C. (2006) Reconciling niche and

721 neutrality: The continuum hypothesis. Ecology Letters, 9, 399–409.

722 Grönroos, M., Heino, J., Siqueira, T., Landeiro, V. L., Kotanen, J., & Bini, L. M. (2013)

723 Metacommunity structuring in stream networks: Roles of dispersal mode, distance

724 type, and regional environmnental context. Ecology and Evolution, 3, 4473–4487.

725 Guisan, A., & Zimmermann, N.E. (2000) Predictive habitat distribution models in ecology.

726 Ecological Modelling, 135, 147–186.

727 Hanski, I. (1994) A practical model of metapopulation dynamics. Journal of Animal Ecology,

728 63, 151–162.

729 Heino, J. (2005) Positive relationship between regional distribution and local abundance in

730 stream insects: a consequence of niche breadth or niche position? Ecography, 28,

731 345–354. 732 Heino, J., & de Mendoza, G. (2016) Predictability of stream insect distributions is dependent

733 on niche position, but not on biological traits or taxonomic relatedness of species.

734 Ecography, 39, 1216–1226.

735 Heino, J., & Grönroos, M. (2014) Untangling the relationships among regional occupancy,

736 species traits and niche characteristics in stream invertebrates. Ecology and Evolution,

737 4, 1931–1942.

738 Heino, J., & Grönroos, M. (2017) Exploring species and site contributions to beta diversity in

739 stream insect assemblages. Oecologia, 183, 151–160.

740 Heino, J., Ilmonen, J., & Paasivirta, L. (2014) Continuous variation of macroinvertebrate

741 communities along environmental gradients in northern streams. Boreal

742 Environment Research, 19, 21–38.

743 Heino, J., Melo, A. S., Siqueira, T., Soininen, J., Valanko, S., & Bini, L. M. (2015)

744 Metacommunity organisation, spatial extent and dispersal in aquatic systems:

745 Patterns, processes and prospects. Freshwater Biology, 60, 845–869.

746 Hoffsten, P.-O. (2004) Site-occupancy in relation to flight-morphology in caddisflies.

747 Freshwater Biology, 49, 810–817.

748 Hubbell, S. P. (2001) The Unified Neutral Theory of Biodiversity and Biogeography.

749 Princeton, NJ: Princeton University Press.

750 Hughes, J. M., Huey, J. A. & Schmidt, D. J. (2013) Is realised connectivity among

751 populations of aquatic fauna predictable from potential connectivity? Freshwater

752 Biology, 58, 951–966.

753 Hughes, J. M., Schmidt, D. J., & Finn, D.S. (2009) Genes in streams: Using DNA to

754 understand the movement of freshwater fauna and their riverine habitat. BioScience,

755 59, 573–583. 756 Ilmonen, J. (2014) Checklist of the family Simuliidae (Diptera) of Finland. ZooKeys, 441,

757 91–95.

758 Jamoneau, A., Passy, S. I., Soininen, J., Lebuocher, T., & Tison-Rosebery, J. (2017) Beta

759 diversity of diatom species and ecological guilds: Response to environmental and

760 spatial mechanisms along the stream watercourse. Freshwater Biology, 63, 62–73.

761 https://doi.org/10.1111/fwb.12980

762 Kaelin, K., & Altermatt, F. (2016) Landscape-level predictions of diversity in river networks

763 reveal opposing patterns for different groups of macroinvertebrates. Aquatic

764 Ecology, 50, 283–295.

765 Kärnä, O.-M., Grönroos, M., Antikainen, H., Hjort, J., Ilmonen, J., Paasivirta, L., & Heino, J.

766 (2015) Inferring the effects of potential dispersal routes on the metacommunity

767 structure of stream insects: as the crow flies, as the fish swims or as the fox runs?

768 Journal of Animal Ecology, 84, 1342–1353.

769 Legendre, P., & Legendre L. (2012) Numerical Ecology (3rd English edition). Elsevier,

770 Amsterdam.

771 Legendre, P., Borcard, D., Blanchet, F.G., & Dray, S. (2013). PCNM: MEM spatial

772 eigenfunction and principal coordinate analyses. R package version 2.1-2/r109.

773 Retrieved from http://R-Forge.R-project.org/projects/sedar/

774 Leibold, M. A. (1995) The niche concept revisited: mechanistic models and community

775 context. Ecology, 76, 1371–1382.

776 Leibold, M. A., Holyoak, M., Mouquet, N., Amarasekare, P., Chase, J. M., Hoopes, M. F.,

777 Holt, R. D., Shurin, J. B., Law, R., Tilman, D., Loreau, M., & Gonzalez, A. (2004)

778 The metacommunity concept: a framework for multi-scale community ecology.

779 Ecology Letters, 7, 601–613. 780 Logue, J. B., Mouquet, N., Peter, H., & Hillebrand, H. (2011) Empirical approaches to

781 metacommunities: a review and comparison with theory. Trends in Ecology and

782 Evolution, 26, 482–491.

783 Maechler, M., Rousseeuw, P., Struyf, A., Hubert, M., & Hornik, K. (2013) cluster: Cluster

784 analysis basics and extensions. R package version 1.14.4. Retrieved from

785 http://CRAN.R-project.org/package=cluster.

786 Mansikkaniemi, H. (1970) Deposits of sorted material in the Inarijoki-Tana river valley in

787 Lapland. Reports of Kevo Subarctic Research Station, 6, 1–63.

788 McCreadie, J. W., & Adler, P. H. (2012) The roles of abiotic factors, dispersal, and species

789 interactions in structuring stream assemblages of black flies (Diptera: Simuliidae).

790 Aquatic Biosystems, 8, 14.

791 Merritt, R. W., & Cummins, K. W. (1996) An Introduction to the Aquatic Insects of North

792 America, 3rd ed. Dubuque: Kendall/Hunt Publishing.

793 Morin, P. J. (2011) Community Ecology, 2nd ed. Oxford: Wiley-Blackwell.

794 Müller-Peddinghaus, E. (2011) Flight-morphology of Central European caddisflies (Insecta:

795 Trichoptera) in relation to their ecological preferences. PhD Thesis, University of

796 Duisburg-Essen, Essen.

797 Müller-Peddinghaus, E., & Hering, D. (2013) The wing morphology of limnephilid

798 caddisflies in relation to their habitat preferences. Freshwater Biology, 58, 1138–

799 1148.

800 Muneepeerakul, R., Bertuzzo, E., Lynch, H. J., Fagan, W. F., Rinaldo, A., & Rodriguez-

801 Iturbe, I. (2008) Neutral metacommunity models predict fish diversity patterns in

802 Mississippi-Missouri basin. Nature, 453, 220–222. 803 National Board of Waters (1981) Vesihallinnon analyysimenetelmät, Tiedotus 213. Helsinki:

804 Vesihallitus.

805 Oksanen, J., Blanchet, F. G., Kindt, R., Legendre, P., Minchin, P. R., O’Hara, R. B.,

806 Simpson, G. L., Solymos, P., Stevens, M. H. H., & Wagner, H. (2013) vegan:

807 Community Ecology Package. R package version 2.0-9. Retrieved from

808 http://CRAN.R-project.org/package=vegan.

809 Pearce, J., & Ferrier, S. (2000) An evaluation of alternative algorithms for fitting species

810 distribution models using logistic regression. Ecological Modelling, 128, 127–147.

811 Peckarsky, B. L. (1983) Biotic interactions or abiotic limitations? A model of lotic

812 community structure. In T. D. Fontaine III, & S. M. Bartell (Eds.), Dynamics of Lotic

813 Ecosystems (pp. 303–323). Ann Arbor: Ann Arbor Science Publishers.

814 Pigot, A. L., & Tobias, J. A. (2013) Species interactions constrain geographic range

815 expansion over evolutionary time. Ecology Letters, 16, 330–338.

816 Pirinen, P., Simola, H., Aalto, J., Kaukoranta, J. P., Karlsson, P., & Ruuhela, R. (2012)

817 Climatological statistics of Finland 1981–2010. Finnish Meteorological Institute

818 Reports, 1, 1–96.

819 Pulido, C., Riera, J. L., Ballesteros, E., Chappuis, E., & Gacia, E. (2015) Predicting aquatic

820 macrophyte occurrence in soft-water oligotrophic lakes (Pyrenees mountain range).

821 Journal of Limnology, 74, 143–154.

822 Pulliam, H. R. (1988) Sources, sinks, and population regulation. American Naturalist, 132,

823 652–661.

824 R Core Team (2013) R: A language and environment for statistical computing. R Foundation

825 for Statistical Computing, Vienna, Austria. Retrieved from http://www.R-project.org/.

826 Schmera, D., Árva, D., Boda, P., Bódis, E., Bolgovics, Á., Borics, G., Csercsa, A., Deák, C.,

827 Krasznai, E. Á., Lukács, B. A., Mauchart, P., Móra, P., Sály, P., Specziár, A., Süveges 828 K., Szivák, I., Takács, P., Tóth, M., Várbíró, G., Vojtkó, A. E., & Erős, T. (2017).

829 Does isolation influence the relative role of environmental and dispersal-related

830 processes in stream networks? An empirical test of the network position hypothesis

831 using multiple taxa. Freshwater Biology, 63, 74–85.

832 https://doi.org/10.1111/fwb.12973

833 Schmidt-Kloiber, A., & Hering, D. (2015) www.freshwaterecology.info – An online tool that

834 unifies, standardises and codifies more than 20,000 European freshwater organisms

835 and their ecological preferences. Ecological Indicators, 53, 271–282.

836 Serra, S. R. Q., Cobo, F., Graça, M. A. S., Dolédec, S., & Feio M. J. (2016) Synthesising the

837 trait information of European Chironomidae (Insecta: Diptera): Towards a new

838 database. Ecological Indicators, 61, 282–292.

839 Sexton, J. P., McIntyre, P. J., Angert, A. L., & Rice, K. J. (2009) Evolution and ecology of

840 species range limits. Annual Review of Ecology, Evolution, and Systematics, 40, 415–

841 436.

842 Seymour, M., Fronhofer, E. A., & Altermatt, F. (2015) Dendritic network structure and

843 dispersal affect temporal dynamics of diversity and species persistence. Oikos, 124,

844 908–916.

845 Southwood, T. R. E. (1977) Habitat, the templet for ecological strategies? Journal of Animal

846 Ecology, 46, 337–365.

847 Tachet, H., Richoux, P., Bournaud, M., & Usseglio-Polatera, P. (2010) Invertébrés d’eau

848 douce: systématique, biologie, écologie (nouvelle édition revue et augmentée).

849 Paris: CNRS.

850 Thomson, J. R., Lake, P. S., & Downes, B. J. (2002) The effect of hydrological disturbance

851 on the impact of a benthic invertertebrate predator. Ecology, 83, 628–642. 852 Vannote, R. L., Minshall, G. W., Cummins, K. W., Sedell, J. R., & Cushing, C.E. (1980) The

853 river continuum concept. Canadian Journal of Fisheries and Aquatic Sciences, 37,

854 130–137.

855 Wallace, J. B., & Webster, J. R. (1996) The role of macroinvertebrates in stream ecosystem

856 function. Annual Review of Entomology, 41, 115–139.

857 Wentworth, C. K. (1922) A scale of grade and class terms for clastic sediments. Journal of

858 Geology, 30, 377–392.

859 Zar, J. H. (1984) Biostatistical Analysis, Second Edition. Prentice Hall, Englewood Cliffs.

860 861 Supporting Information

862 Additional Supporting Information may be found in the online version of this article.

863 Table S1. Insect species considered.

864 Table S2. Species traits considered and site occupancy.

865 Table S3. Results of binomial GLMs.

866 Table S4. Taxonomic and trait vectors from Principal Coordinate Analysis (PCO), with

867 corresponding scores for each species.

868 Table S5. Statistical significance of the correlations among site occupancy, species traits,

869 trait vectors, and taxonomic vectors.

870 Figure S1. A map of the study area located in the Tenojoki drainage basin.

871 Figure S2. Eigenvalues from taxonomic Principal Coordinate Analysis (PCO) based on

872 taxonomic distances between species.

873 Figure S3. Results of binomial GLMs in relation to site occupancy and species traits.

874 Figure S4. Comparison of site occupancy values between different insect groups.

875 876 Table 1. Results of beta regression showing the effects of site occupancy, biological trait

877 vectors and taxonomic vectors on different fractions of variation (adjusted deviance, Adj. D2)

878 explained by binomial GLMs: local environment (E) effects, climate (C) effects, spatial (S)

879 effects, and combined (E+C+S) effects. Significant values (p < .05) are shown in boldface;

880 marginally significant values (p < .10) in italics.

Adj. D2 of E effects Estimate SE z p Log- Pseudo R2 likelihood (Intercept) -1.6002 0.2220 -7.209 <0.001 49.06 0.2500 Site occupancy -0.0042 0.0091 -0.469 0.646 TAX-PCO1 -378.2072 323.7056 -1.168 0.243 TAX-PCO2 91.4160 161.7362 0.565 0.572 TAX-PCO3 225.8850 218.0393 1.036 0.300 TAX-PCO4 -112.1618 101.9824 -1.100 0.271 TRA-PCO1 -0.5557 0.8628 -0.644 0.520 TRA-PCO2 1.3525 1.0712 1.263 0.207 TRA-PCO3 0.3959 0.9475 0.418 0.676 TRA-PCO4 -1.3110 1.3459 -0.974 0.330

Adj. D2 of C effects Estimate SE z p Log- Pseudo R2 likelihood (Intercept) -2.9806 0.2682 -11.115 <0.001 94.66 0.2309 Site occupancy 0.0029 0.0103 0.278 0.781 TAX-PCO1 -579.9412 375.5543 -1.544 0.123 TAX-PCO2 83.1028 185.9703 0.447 0.655 TAX-PCO3 352.6843 252.8857 1.395 0.163 TAX-PCO4 -1.9501 120.4650 -0.016 0.987 TRA-PCO1 -1.6362 1.0045 -1.629 0.103 TRA-PCO2 0.4327 1.2476 0.347 0.729 TRA-PCO3 0.6454 1.1129 0.580 0.562 TRA-PCO4 -2.1768 1.5519 -1.403 0.161

Adj. D2 of S effects Estimate SE z p Log- Pseudo R2 likelihood (Intercept) -2.4436 0.2222 -10.997 <0.001 60.19 0.2137 Site occupancy 0.0208 0.0083 2.498 0.012 TAX-PCO1 309.6453 360.2543 0.860 0.390 TAX-PCO2 -4.0452 162.3764 -0.025 0.980 TAX-PCO3 365.5554 205.9633 1.775 0.076 TAX-PCO4 217.5296 102.0912 2.131 0.033 TRA-PCO1 -1.0051 0.8333 -1.206 0.228 TRA-PCO2 2.5228 1.1999 2.102 0.036 TRA-PCO3 1.4151 0.9765 1.449 0.147 TRA-PCO4 -0.7984 1.3342 -0.598 0.550

Adj. D2 of E+C+S effects Estimate SE z p Log- Pseudo R2 likelihood (Intercept) -0.7133 0.2331 -3.060 0.002 17.76 0.1791 Site occupancy 0.0111 0.0095 1.164 0.244 TAX-PCO1 -140.2227 352.1883 -0.398 0.691 TAX-PCO2 32.5012 178.6572 0.182 0.856 TAX-PCO3 407.2860 228.4095 1.783 0.075 TAX-PCO4 81.0016 110.2921 0.734 0.463 TRA-PCO1 -1.3052 0.9142 -1.428 0.153 TRA-PCO2 1.6446 1.1935 1.378 0.168 TRA-PCO3 0.9742 1.0367 0.940 0.347 TRA-PCO4 -1.0409 1.4575 -0.714 0.475 881 882 Figure legends

883 Figure 1 Conceptual representation of (a) the four non-exclusive classical approaches in

884 metacommunity studies (Leibold et al. 2004), and (b) the more recent framework of

885 metacommunity analysis based on three exclusive components (Logue et al. 2011); according

886 to the relative relevance of the variables used in this study (axes): spatial variables (x-axis),

887 environmental variables (y-axis) and the different dispersal capability of species (z-axis).

888 Circles represent the theoretical location where the emphasis of each approach is situated

889 across the three axes.

890 Figure 2 Flow chart of the statistical analyses performed in this study.

891 Figure 3 Frequency of local environmental variables (top) and spatial variables (bottom)

892 selected as significant in explaining species distributions through binomial GLMs. Spatial

893 variables are arranged from small-scale (i.e. V13) to large-scale extent (i.e. V1), and include

894 V11 which was never selected. Climate variables are not shown (mean annual temperature

895 was selected 32 times, and July air temperature 20 times). Species-environment relationships

896 are shown in black when positive and grey when negative. Specific information for each

897 species can be found in Table S2.

898 Figure 4 Species scores on taxonomic vectors TAX-PCO3 (a) and TAX-PCO4 (b), and on

899 trait vector TRA-PCO2 (c), arranged from lowest to highest values. For the trait vector, the

900 location of the different functional feeding (FFG) and habit trait groups (HTG), is indicated,

901 as well as the four different body size classes considered (BS, represented by columns of four

902 different sizes), and those insects considered as of high female dispersal potential (H).

903 904 Figure 1

905 a)

Species Patch sorting dynamics

Source-sink dynamics

Neutral

theory Environmental variables Environmental

Spatial variables

b)

Environmental heterogeneity

Dispersal

Species

equivalence Environmental variables Environmental

Spatial variables 906 Figure 2

907 908 Figure 3

909 Bank height Velocity Local environmental variables Sand Depth Nitrogen Manganese Bank slope Pebble Moss Cobble Stream width Shading Water temperature

V13 Spatial variables V12 V11 V10 V9 V8 V7 V6 V5 V4 V3 V2 V1

0 2 4 6 8 10 12 14

Frequency 910 Figure 4

0.0035 Simuliidae 0.0030 a 0.0025

0.0020 0.0015 Coleoptera 0.0010

PCO3 score Ephemeroptera Chironomidae Trichoptera Plecoptera

- 0.0005 0.0000

TAX -0.0005 -0.0010 -0.0015 0.0040 b Trichoptera 0.0030

0.0020 Ephemeroptera Coleoptera

0.0010 Simuliidae Plecoptera Chironomidae

0.0000

PCO4 score

-

-0.0010

TAX

-0.0020

-0.0030 0.5000 c Ephemeroptera 0.4000 T T Plecoptera 0.3000 Trichoptera Coleoptera 0.2000 T E Chironomidae P P 0.1000 Simuliidae

PCO2 score - 0.0000

Shredders Predators Scrapers TRA -0.1000 Predators Shredders FFG -0.2000 Scrapers Gathering-collectors -0.3000 Shredders Filtering-collectors Filtering-collectors Cl.

Clingers Sprawlers Clingers Clingers Clingers Swimmers HTG Sprawlers Sprawlers Sprawlers BS H H H H H H H H DP Supporting Information

Highly variable species distribution models in a subarctic stream metacommunity: patterns, mechanisms and implications

Guillermo de Mendoza, Riikka Kaivosoja, Mira Grönroos, Jan Hjort, Jari Ilmonen,

Olli-Matti Kärnä, Lauri Paasivirta, Laura Tokola & Jani Heino

Freshwater Biology

Table S1. Insect species considered.

Table S2. Species traits considered and site occupancy.

Table S3. Results of binomial GLMs.

Table S4. Taxonomic and trait vectors from Principal Coordinate Analysis (PCO), with corresponding scores for each species.

Table S5. Statistical significance of the correlations among site occupancy, species traits, trait vectors, and taxonomic vectors.

Figure S1. A map of the study area located in the Tenojoki drainage basin.

Figure S2. Eigenvalues from taxonomic Principal Coordinate Analysis (PCO) based on taxonomic distances between species.

Figure S3. Results of binomial GLMs in relation to site occupancy and species traits.

Figure S4. Comparison of site occupancy values between different insect groups. Table S1. Insect species considered, ordered alphabetically, with information about their higher classification (order, family) and codes used.

Species Higher classification Code Ameletus inopinatus Eaton Ephemeroptera, Ameletidae Ame.ino Amphinemura borealis (Morton) Plecoptera, Nemouridae Amp.bor Amphinemura sulcicollis (Stephens) Plecoptera, Nemouridae Amp.sul Apatania muliebris McLachlan Trichoptera, Apataniidae Apa.mul Baetis muticus (Linnaeus) Ephemeroptera, Baetidae Bae.mut Baetis cf. niger (Linnaeus) Ephemeroptera, Baetidae Bae.nig Brachyptera risi (Morton) Plecoptera, Taeniopterygidae Bra.ris Cardiocladius capucinus (Zetterstedt) Diptera, Chironomidae Car.cap Cardiocladius fuscus Kieffer Diptera, Chironomidae Car.fus Constempellina brevicosta (Edwards) Diptera, Chironomidae Con.bre Corynoneura lobata-type Diptera, Chironomidae Cor.lob Diura nanseni (Kempny) Plecoptera, Perlodidae Diu.nan Elmis aenea (Muller) Coleoptera, Elmidae Elm.aen Ephemerella aroni Eaton Ephemeroptera, Ephemerellidae Eph.aro Eukiefferiella boevrensis Brundin Diptera, Chironomidae Euk.boe Eukiefferiella brevicalcar (Kieffer) Diptera, Chironomidae Euk.bre Eukiefferiella devonica-group Diptera, Chironomidae Euk.dev Helodon ferrugineus (Wahlberg) Diptera, Simuliidae Hel.fer Heptagenia dalecarlica Bengtsson Ephemeroptera, Heptageniidae Hep.dal Hydraena gracilis Germar Coleoptera, Hydraenidae Hyd.gra Isoperla difformis (Klapalek) Plecoptera, Perlodidae Iso.dif Leuctra cf. digitata Kempny Plecoptera, Leuctridae Leu.dig Leuctra nigra (Olivier) Plecoptera, Leuctridae Leu.nig Metacnephia bilineata (Rubtsov) Diptera, Simuliidae Met.bil Micropsectra atrofasciata-group Diptera, Chironomidae Mic.atr Orthocladuis excavatus-type Diptera, Chironomidae Ort.exc Orthocladius frigidus (Zetterstedt) Diptera, Chironomidae Ort.fri Orthocladius olivaceus-type Diptera, Chironomidae Ort.oli Orthocladius rhyacobius-group Diptera, Chironomidae Ort.rhy Orthocladius rivicola Kieffer Diptera, Chironomidae Ort.riv Orthocladius rivulorum Kieffer Diptera, Chironomidae Ort.riu Paratrichocladius skirwithensis (Edwards) Diptera, Chironomidae Par.ski Philopotamus montanus (Donovan) Trichoptera, Philopotamidae Phi.mon Plectrocnemia conspersa (Curtis) Trichoptera, Polycentropodidae Ple.con Potamophylax cf. cingulatus (Stephens) Trichoptera, Limnephilidae Pot.cin Potthastia longimana Kieffer Diptera, Chironomidae Pot.lon Prosimulium hirtipes (Fries) Diptera, Simuliidae Pro.hir Rhyacophila nubila Zetterstedt Trichoptera, Rhyacophilidae Rhy.nub Simulium monticola Friederichs Diptera, Simuliidae Sim.mon Simulium murmanum Enderlein Diptera, Simuliidae Sim.mur Siphonoperla burmeisteri (Pictet) Plecoptera, Chloroperlidae Sip.bur Stegopterna trigonium (Lundstrom) Diptera, Simuliidae Ste.tri Thienemanniella majuscula-type Diptera, Chironomidae Thi.maj Trissopelopia longimana (Staeger) Diptera, Chironomidae Tri.lon Tvetenia bavarica (Goetghebuer) Diptera, Chironomidae Tve.bav Tvetenia calvescens (Edwards) Diptera, Chironomidae Tve.cal Tvetenia discoloripes (Goetghebuer & Thienemann) Diptera, Chironomidae Tve.dis Table S2. Species traits considered and site occupancy (sites). Body size class (cm) refers to maximum size. Species ordered alphabetically by their codes (Table S1).

Species Sites Body size Dispersal Functional feeding group Habit trait group Ame.ino 33 1-2 low Scraper Swimmer Amp.bor 23 0.5-1 low Shredder Sprawler Amp.sul 28 0.5-1 low Shredder Sprawler Apa.mul 6 1-2 low Scraper Sprawler Bae.mut 38 1-2 low Scraper Swimmer Bae.nig 24 1-2 low Scraper Swimmer Bra.ris 24 1-2 low Scraper Sprawler Car.cap 6 0.5-1 low Predator Clinger Car.fus 11 0.5-1 low Predator Clinger Con.bre 6 0-0.5 low Gathering collector Clinger Cor.lob 12 0-0.5 low Gathering collector Sprawler Diu.nan 10 2-4 low Predator Clinger Elm.aen 14 0-0.5 low Scraper Clinger Eph.aro 31 0.5-1 low Gathering collector Clinger Euk.boe 8 0-0.5 low Gathering collector Sprawler Euk.bre 8 0-0.5 low Gathering collector Sprawler Euk.dev 33 0-0.5 low Gathering collector Sprawler Hel.fer 41 0.5-1 high Filtering collector Clinger Hep.dal 31 1-2 low Scraper Clinger Hyd.gra 7 0-0.5 low Scraper Clinger Iso.dif 42 1-2 low Predator Clinger Leu.dig 31 1-2 low Shredder Sprawler Leu.nig 20 1-2 low Shredder Sprawler Met.bil 32 0.5-1 high Filtering collector Clinger Mic.atr 43 0.5-1 low Gathering collector Sprawler Ort.exc 7 0.5-1 low Gathering collector Sprawler Ort.fri 26 0.5-1 low Gathering collector Sprawler Ort.oli 12 0.5-1 low Gathering collector Sprawler Ort.rhy 19 0.5-1 low Gathering collector Sprawler Ort.riu 9 0.5-1 low Gathering collector Sprawler Ort.riv 44 0.5-1 low Gathering collector Sprawler Par.ski 9 0.5-1 low Gathering collector Sprawler Phi.mon 18 1-2 low Filtering collector Clinger Ple.con 14 2-4 high Predator Clinger Pot.cin 9 2-4 high Shredder Clinger Pot.lon 11 0.5-1 low Gathering collector Sprawler Pro.hir 45 0.5-1 high Filtering collector Clinger Rhy.nub 51 2-4 low Predator Clinger Sim.mon 41 0.5-1 high Filtering collector Clinger Sim.mur 29 0.5-1 high Filtering collector Clinger Sip.bur 10 1-2 low Predator Clinger Ste.tri 6 0.5-1 high Filtering collector Clinger Thi.maj 9 0-0.5 low Gathering collector Sprawler Tri.lon 14 0.5-1 low Predator Sprawler Tve.bav 15 0.5-1 low Gathering collector Sprawler Tve.cal 31 0.5-1 low Gathering collector Sprawler Tve.dis 17 0.5-1 low Gathering collector Sprawler Table S3. Results of binomial GLMs. Deviance explained (adjusted D2 values) is decomposed into pure local environment (E) effects, pure climate (C) effects, pure spatial (S) effects, and combined (E+C+S) pure and joint effects. The effect (+/-) of environmental variables on species distributions is also shown. Environmental and spatial (V) variables are ordered as selected by each model (climate variables not shown). Species codes are as in Table S1.

Species E C S E+C+S Variables selected in binomial models Ame.ino 11.2 5.5 22.2 41.7 Sand (-), Water temperature (-); V4, V8, V1, V9 Amp.bor 21.6 13.1 5.1 37.1 Stream width (+), Water temperature (+); V6 Amp.sul 10.8 7.4 9.5 43.6 Velocity (+), Cobble (+), Bank slope (+); V1 Apa.mul 31.9 7.0 20.6 74.1 Pebble (+); V1, V8 Bae.mut 24.2 1.8 8.4 40.8 Stream width (+); V2, V4 Bae.nig 12.1 1.1 2.7 17.2 Stream width (+); V2 Bra.ris 22.9 17.0 31.5 70.4 Water temperature (+); V3, V4 Car.cap 9.6 1.0 28.6 42.4 Water temperature (+), Bank slope (+); V1, V5 Car.fus 5.5 1.0 3.1 12.5 Depth (+); V9 Con.bre 25.7 3.4 4.6 46.0 Manganese (+); V6 Cor.lob 9.5 12.6 17.2 45.0 Shading (+), Moss (+); V2, V12 Diu.nan 31.6 34.4 20.2 53.1 Pebble (+), Velocity (+); V4 Elm.aen 8.8 0.7 7.5 16.6 Moss (-); V3 Eph.aro 11.3 3.9 7.4 29.1 Stream width (-), Bank slope (-); V3, V4 Euk.boe 13.9 8.6 0.7 30.8 Sand (+); V6 Euk.bre 26.9 1.2 16.9 53.9 Moss (+), Shading (+); V2 Euk.dev 1.1 5.8 8.0 17.9 Depth (-); V8 Hel.fer 16.8 2.1 13.3 33.2 Cobble (+); V10, V3 Hep.dal 50.2 3.5 13.3 67.2 Stream width (+); V2 Hyd.gra 14.1 0.9 10.5 29.7 Manganese (+); V3 Iso.dif 19.0 1.8 19.1 28.6 Shading (+); V1, V4 Leu.dig 33.6 3.7 2.8 44.3 Water temperature (-), Shading (+), Cobble (+); V9 Leu.nig 17.0 3.7 10.4 29.4 Shading (+); V9 Met.bil 7.5 5.9 10.2 26.9 Water temperature (+); V13, V4 Mic.atr 14.2 1.3 8.8 33.9 Water temperature (-); V7, V1 Ort.exc 5.7 2.4 38.6 60.1 Water temperature (-); V4, V5 Ort.fri 3.0 0.4 2.7 19.4 Cobble (+), Pebble (+); V4 Ort.oli 13.4 0.4 6.9 20.7 Nitrogen (-); V3 Ort.rhy 6.8 12.4 21.7 44.5 Bank height (-), W. temp. (-), Shad. (+); V1, V3, V7 Ort.riu 2.5 0.4 6.9 10.5 Depth (+); V3 Ort.riv 9.1 28.9 23.5 55.5 Nitrogen (-); V10, V12 Par.ski 22.9 1.2 5.1 37.5 Moss (+), Bank slope (-); V7 Phi.mon 3.1 0.7 13.1 21.0 Shading (+); V1, V3 Ple.con 5.6 0.4 5.0 11.0 Shading (+); V1 Pot.cin 13.3 2.1 2.0 21.3 Stream width (-), Moss (+); V2 Pot.lon 6.0 1.1 5.2 15.3 Cobble (+); V3 Pro.hir 37.7 2.2 19.1 91.7 Shading (+), Nitrogen (-), Pebble (+); V9, V12 Rhy.nub 14.4 11.0 41.2 72.2 Pebble (+); V1 Sim.mon 8.8 1.6 9.2 30.2 Nitrogen (-); V3, V8 Sim.mur 12.0 1.8 8.5 22.2 Water temperature (+); V7 Sip.bur 66.7 20.1 1.6 82.1 Stream width (+); V3 Ste.tri 2.0 1.7 6.5 25.0 Water temperature (+); V3 Thi.maj 18.8 2.0 15.4 30.4 Moss (+); V10 Tri.lon 11.0 0.7 12.5 22.0 Manganese (-); V5 Tve.bav 20.8 6.2 2.2 28.1 Moss (+); V12 Tve.cal 16.8 3.3 15.3 33.2 Cobble (+), Manganese (+); V6, V3 Tve.dis 5.8 5.2 28.6 56.0 Shading (+), Cobble (+), Bank height (-); V1, V8 Mean 16.1 5.4 12.6 37.8 Minimum 1.1 0.4 0.7 10.5 Maximum 66.7 34.4 41.2 91.7 Table S4. Taxonomic (TAX) and trait (TRA) vectors from Principal Coordinate Analysis (PCO), with corresponding scores for each species (codes of Table S1 in alphabetical order).

Species TAX-1 TAX-2 TAX-3 TAX-4 TRA-1 TRA-2 TRA-3 TRA-4 Ame.ino -0.000672 -0.001588 -0.000936 -0.002100 -0.0293 0.3933 0.1122 0.1866 Amp.bor -0.000883 0.001971 0.000031 -0.000206 -0.1962 -0.0124 -0.2275 0.0800 Amp.sul -0.000883 0.001971 0.000031 -0.000206 -0.1962 -0.0124 -0.2275 0.0800 Apa.mul -0.000667 -0.001318 -0.000151 0.002842 -0.1949 0.2649 -0.0263 0.1285 Bae.mut -0.000710 -0.001887 -0.001175 -0.002686 -0.0293 0.3933 0.1122 0.1866 Bae.nig -0.000710 -0.001887 -0.001175 -0.002686 -0.0293 0.3933 0.1122 0.1866 Bra.ris -0.000836 0.001659 0.000025 -0.000161 -0.1949 0.2649 -0.0263 0.1285 Car.cap 0.000627 0.000177 -0.000519 0.000067 0.1920 0.0249 0.0730 -0.2107 Car.fus 0.000627 0.000177 -0.000519 0.000067 0.1920 0.0249 0.0730 -0.2107 Con.bre 0.000615 0.000167 -0.000484 0.000062 0.0047 -0.1903 0.2542 -0.1198 Cor.lob 0.000615 0.000167 -0.000484 0.000062 -0.3372 -0.1323 0.0702 -0.0207 Diu.nan -0.000864 0.001841 0.000028 -0.000186 0.2721 0.2469 -0.0171 -0.2712 Elm.aen -0.000564 -0.000560 -0.000025 0.000208 0.1171 0.0981 0.3129 0.0977 Eph.aro -0.000672 -0.001588 -0.000936 -0.002100 0.0407 -0.1447 0.1404 -0.0843 Euk.boe 0.000640 0.000188 -0.000560 0.000073 -0.3372 -0.1323 0.0702 -0.0207 Euk.bre 0.000640 0.000188 -0.000560 0.000073 -0.3372 -0.1323 0.0702 -0.0207 Euk.dev 0.000640 0.000188 -0.000560 0.000073 -0.3372 -0.1323 0.0702 -0.0207 Hel.fer 0.000222 -0.000436 0.002914 -0.000492 0.4740 -0.2543 -0.0050 0.1128 Hep.dal -0.000672 -0.001588 -0.000936 -0.002100 0.1923 0.2710 0.1300 0.1053 Hyd.gra -0.000564 -0.000560 -0.000025 0.000208 0.1171 0.0981 0.3129 0.0977 Iso.dif -0.000864 0.001841 0.000028 -0.000186 0.2259 0.1929 0.0156 -0.1839 Leu.dig -0.000883 0.001971 0.000031 -0.000206 -0.1853 0.1657 -0.3076 0.0424 Leu.nig -0.000883 0.001971 0.000031 -0.000206 -0.1853 0.1657 -0.3076 0.0424 Met.bil 0.000222 -0.000436 0.002914 -0.000492 0.4740 -0.2543 -0.0050 0.1128 Mic.atr 0.000615 0.000167 -0.000484 0.000062 -0.2591 -0.0955 -0.0218 -0.0117 Ort.exc 0.000681 0.000230 -0.000732 0.000097 -0.2591 -0.0955 -0.0218 -0.0117 Ort.fri 0.000681 0.000230 -0.000732 0.000097 -0.2591 -0.0955 -0.0218 -0.0117 Ort.oli 0.000681 0.000230 -0.000732 0.000097 -0.2591 -0.0955 -0.0218 -0.0117 Ort.rhy 0.000681 0.000230 -0.000732 0.000097 -0.2591 -0.0955 -0.0218 -0.0117 Ort.riu 0.000681 0.000230 -0.000732 0.000097 -0.2591 -0.0955 -0.0218 -0.0117 Ort.riv 0.000681 0.000230 -0.000732 0.000097 -0.2591 -0.0955 -0.0218 -0.0117 Par.ski 0.000615 0.000167 -0.000484 0.000062 -0.2591 -0.0955 -0.0218 -0.0117 Phi.mon -0.000667 -0.001318 -0.000151 0.002842 0.2777 0.0496 0.0362 0.0862 Ple.con -0.000667 -0.001318 -0.000151 0.002842 0.6379 0.0928 -0.1271 -0.2234 Pot.cin -0.000667 -0.001318 -0.000151 0.002842 0.5730 0.0584 -0.3733 0.1310 Pot.lon 0.000615 0.000167 -0.000484 0.000062 -0.2591 -0.0955 -0.0218 -0.0117 Pro.hir 0.000222 -0.000436 0.002914 -0.000492 0.4740 -0.2543 -0.0050 0.1128 Rhy.nub -0.000667 -0.001318 -0.000151 0.002842 0.2721 0.2469 -0.0171 -0.2712 Sim.mon 0.000226 -0.000460 0.003126 -0.000531 0.4740 -0.2543 -0.0050 0.1128 Sim.mur 0.000226 -0.000460 0.003126 -0.000531 0.4740 -0.2543 -0.0050 0.1128 Sip.bur -0.000836 0.001659 0.000025 -0.000161 0.2259 0.1929 0.0156 -0.1839 Ste.tri 0.000222 -0.000436 0.002914 -0.000492 0.4740 -0.2543 -0.0050 0.1128 Thi.maj 0.000615 0.000167 -0.000484 0.000062 -0.3372 -0.1323 0.0702 -0.0207 Tri.lon 0.000615 0.000167 -0.000484 0.000062 -0.1479 0.0545 -0.1030 -0.2533 Tve.bav 0.000640 0.000188 -0.000560 0.000073 -0.2591 -0.0955 -0.0218 -0.0117 Tve.cal 0.000640 0.000188 -0.000560 0.000073 -0.2591 -0.0955 -0.0218 -0.0117 Tve.dis 0.000640 0.000188 -0.000560 0.000073 -0.2591 -0.0955 -0.0218 -0.0117

Table S5. Statistical significance (P-values) of the correlation between site occupancy, species traits, trait vectors, and taxonomic vectors. Depending on the continuous or categorical nature of the variables involved, P-values refer to either the significance of the Spearman correlation coefficient, to Kruskal-Wallis tests, to Mann-Whitney tests, or to Fisher’s exact tests on a contingency table. Although P-values from these various analyses are not strictly comparable, they provide an overall idea of the degree to which two variables are correlated. Nsites, site occupancy; BS, body size; DP, dispersal potential; FFG, functional feeding group; HTG, habit trait group; TRA, trait vector; TAX, taxonomic vector. Vectors are multivariate axes of a Principal Coordinate Analysis combining either trait or taxonomic variables. Significant P-values (P < 0.05) are shown in boldface; marginally significant (P < 0.10) in italics.

Nsites BS DP FFG HTG TAX-1 TAX-2 TAX-3 TAX-4 TRA-1 TRA-2 TRA-3 TRA-4 Nsites - BS 0.108 - DP 0.275 0.043 - FFG 0.468 <0.001 <0.001 - HTG 0.275 0.024 0.001 <0.001 - TAX-1 0.112 <0.001 0.599 <0.001 0.009 - TAX-2 0.547 0.492 0.019 <0.001 <0.001 0.376 - TAX-3 0.637 0.750 <0.001 <0.001 <0.001 <0.001 0.373 - TAX-4 0.007 0.022 0.151 0.072 0.008 <0.001 0.496 0.223 - TRA-1 0.224 0.004 <0.001 <0.001 <0.001 <0.001 0.002 <0.001 0.083 - TRA-2 0.808 <0.001 0.003 <0.001 0.015 <0.001 0.793 0.220 0.591 0.337 - TRA-3 0.784 0.002 0.467 0.002 <0.001 0.869 <0.001 0.085 0.103 0.382 0.626 - TRA-4 0.101 0.028 0.014 <0.001 0.015 0.038 0.011 0.250 0.019 0.120 0.586 0.783 -

Fig. S1. A map of the study area located in the Tenokoki drainage basin. Both the location of the Tenojoki basin at continental scale (A), and the details of the river network (B), are shown. All sampling points are located in tributary streams of the River Tenojoki or the River Utsjoki. Dashed lines show the border between Norway and Finland. Note that only parts of the stream networks on the Norwegian side of the border are shown for clarity.

Fig. S2. Eigenvalues from taxonomic Principal Coordinate Analysis (PCO) based on taxonomic distances between species. The first four eigenvectors were selected in the comparative analysis across species.

50000

40000

30000

20000 Eigenvalue 10000

0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47

# of eigenvector

Fig. S3 (see next page). Results of binomial GLMs (adjusted D2 values) in relation to site occupancy and species traits. From left to right, columns refer to site occupancy, body size classes (cm), dispersal potential, functional feeding groups, and habit trait groups, respectively. The adjusted D2 values of binomial GLMs, result from pure local environmental (E), pure climate (C), or pure spatial (S) effects, as well as from a combination of the pure and joint effects of the three subsets of variables (E+C+S). Numbers inside graphs refer to P- values of univariate effects of variables on the deviance explained: beta regressions for site occupancy; Kruskal-Wallis tests for body size classes, functional feeding groups, and habit trait groups; and Mann-Whitney tests for dispersal potential. Significance (P < 0.05) is indicated with an asterisk. Letters inside graphs indicate significant differences (P < 0.05, Mann-Whitney tests) for pairwise comparisons, in case of statistical significance of the Kruskal-Wallis test. Abbreviations: Filt, filtering collector; Gath, gathering collector, Pred, predator; Scrap, scraper; Shred, shredder; Clin, clinger; Spra, sprawler; Swim, swimmer. Fig. S3

Fig. S4. Comparison of site occupancy values between different insect groups. a) broad taxonomic groups; b) functional feeding groups; c) habit trait groups. Numbers inside boxes indicate the P-value of the Kruskal-Wallis test of each analysis, with the asterisk indicating marginal statistical significance (P < 0.10). In case the Kruskal-Wallis test was (marginally) significant, letters inside boxes indicate marginal statistical differences (P < 0.10; Mann- Whitney test; in a) the Simuliidae differed from the Chironomidae but not from the other groups).

a) 55 a a ab ab ab b 50 45 40 35 30 25 20 Site occupancy Site 15 10 5 P = 0.057* 0

Coleoptera Trichoptera Plecoptera Simuliidae Chironomidae Ephemeroptera

b) 55 50 45 40 35 30 25 20 Site occupancy Site 15 10 5 P = 0.468 0

Scrapers Shredders Predators

Filtering-collectors Gathering-collectors c) 55 50 45 40 35 30 25 20 Site occupancy Site 15 10 5 P = 0.275 0

Clingers Sprawlers Swimmers